Classification by using BbNN

1
Classification by using BbNN

• Sang Ki Park
• July.12.2005

2
Contents

• 2D BbNN 3D BbNN
• Extend 2D BbNN to 3 D BbNN model to obtaining
• more connections for increase
  classification performance and fault tolerance.

3
2D BbNN 3D BbNN

• Motivation
• Need more connection and diversity for increasing
  classification performance and fault tolerance
• Give more connection within the network has same
  number of block
• The distances between Inputs are different in 2D
  BbNN
• Need more path in Block

x
x
w1
x
w3
?
w2
n3
w4
n1
n2
n3
4
2D BbNN 3D BbNN

• 3D BbNN Model
• 2 x 1bit connections and 16 x 8bit weights
  (include biases) in a block

Combine two 2D Blocks
5
2D BbNN 3D BbNN

• 2D 3D BbNN Configurations Comparison

6
2D BbNN 3D BbNN

• The Connections and the Distance between Blocks
  in BbNN models

Configure (Row x Column x Layer)
7
2D BbNN 3D BbNN

• Iris Classification Result
• 4input, 2output (3 class), 75 Training data, 75
  Test data (Total 150data)

8
2D BbNN 3D BbNN

• Wisconsin Breast-Cancer Classification Result
• 9input, 1output (2class), 75 Training data, 624
  Test data (Total 699 data)

9
Conclusion

• 3D BbNN shows better performance in
  classification test than 2D BbNN.
• In Same Block Sized Network, 3D BbNN can give
  more connections.
• From the Results of the Classification Tests, 3D
  BbNN shows better performance.
• In the contrary, 3D model need more memory than
  2D model.
• Developing 2D and 3D model to reduce the number
  of block size for solving ECG Classification
  problem and Feature Extraction method